Data inconsistencies are expected, not exceptional

Data Trust Explorer

bp Sphere does not assume perfect data. It expects timing differences, missing attributes, duplicate records, conflicting values, and hybrid ERP landscapes.
Trust is operationalized through source-system lineage, live/simulated mode, evidence completeness, conflict detection, policy authority, decision replay, collaboration handoff, action boundary, observability, and value impact for every finance case.
Provenance: reference trust case, not live conflict telemetry. Decision IDs, supplier names, confidence values, and conflicts are illustrative until backed by live semantic-conflict records and SOR lineage.
Decision
INV-100245
Hold for Review
Supplier
Shell Services
$245,000
Decision Confidence
92%
Recommend + human review band
Data Trust Architecture
Layer 1 - Source DataSAP, ECC, S4, CFIN, UDP, external and manual sources
v
Layer 2 - Normalizationmapping, cleansing, deduplication, canonical linking
v
Layer 3 - Business Contextontology, process meaning, ownership, relationship resolution
v
Layer 4 - Confidence Enginetrust score, conflict score, freshness, source agreement
v
Layer 5 - Agent Decisioninggoverned action, human review, escalation, or block
Trust Scores
Decision Confidence92%
Data Quality Score88%
Completeness90%
Source Agreement84%
Policy Alignment95%
Source Systems
SourceStatusFreshnessConfidence
SAP ECCAvailableLive validation98
SAP S/4Available5 minutes96
CFINAvailable7 minutes94
DatabricksAvailable5 minutes95
Contract RepositoryAvailable1 hour90
Missing Fields
FieldImpactRemediation
Payment TermsLowrefresh vendor profile before payment execution
Tax ClassificationMediumroute to tax validation before autonomous action
Conflicts Detected
AttributeSource ASource BResolution
Supplier CategorySAP ECC: StrategicAriba: Standardprefer semantic authority; require steward review for payment-term action
Cost CenterCFIN: 2100S/4: 2105allow recommendation only; block write-back until owner confirms
Confidence-Driven Agent Behavior
ConfidenceAgent BehaviorControl
>95%Autonomous Actionpolicy and evidence gates still required
80-95%Recommend + Human Reviewsupervisor approval before execution
60-80%Escalaterequest additional context or steward remediation
<60%Block Decisionno recommendation promoted to action
Core Principle
  • Data is available
  • Context is understood
  • Confidence is sufficient
  • Policy permits execution
Runtime Support
  • Semantic Conflicts Table: iaf_semantic_conflicts
  • Canonical Identity Binding: iaf_canonical_identity_bindings
  • Source Lineage Policy: source-lineage-policy
  • Evidence Completeness Policy: evidence-completeness-policy
  • Hybrid Finance Replication Gap: True
  • Confidence Driven Autonomy: True
  • Stale Or Conflicting Data Blocks Writeback: True
Message for BP Technology Leadership
Data quality is not treated as a prerequisite for automation. Data quality becomes observable, measurable, and governable inside every decision.